| OUTPUT_DIR='/ccn2/u/honglinc/cwm_checkpoints/ablation_3frame_clumping_v2/' | |
| DATA_PATH="${HOME}/BBNet/bbnet/models/VideoMAE-main/video_file_lists/kinetics_400_train_list.txt" | |
| CUDA_VISIBLE_DEVICES=0 OMP_NUM_THREADS=1 python -m torch.distributed.launch --nproc_per_node=1 \ | |
| --master_addr=10.102.2.157 --master_port=32240 \ | |
| --nnodes=1 --node_rank=0 \ | |
| run_cwm_pretraining.py \ | |
| --data_path ${DATA_PATH} \ | |
| --mask_type rotated_table \ | |
| --mask_ratio 0.99 \ | |
| --mask_kwargs '{"tube_length": 1}' \ | |
| --model vitbase_8x8patch_3frames_1tube_cf2 \ | |
| --context_frames 2 \ | |
| --target_frames 1 \ | |
| --temporal_units 'ms' \ | |
| --sampling_rate 150 \ | |
| --context_target_gap 150 150 \ | |
| --batch_size 3 \ | |
| --opt adamw \ | |
| --opt_betas 0.9 0.95 \ | |
| --warmup_epochs 40 \ | |
| --save_ckpt_freq 10 \ | |
| --epochs 800 \ | |
| --no_normlize_target \ | |
| --rescale_size 224 \ | |
| --augmentation_type 'multiscale' \ | |
| --augmentation_scales 1.0 0.875 0.75 0.66 \ | |
| --log_dir ${OUTPUT_DIR} \ | |
| --output_dir ${OUTPUT_DIR} \ | |
| --print_freq 1 \ | |
| --num_workers 0 | |